A method of this kind is described in an article “Achieving ‘Always Best Connected’ through Extensive Profile Management”, presented at the 9th International Conference on Personal Wireless Communications, September 2004. In systems implementing the known method, multiple communications networks enable the terminals to communicate with each other through a plurality of interfaces with which each terminal is provided.
The known method enables detection of the loss of an interface during the operation of a given terminal, and the almost instantaneous rerouting of a stream, to which the lost interface will have been initially assigned, to another interface using a first decision-making algorithm called a fast algorithm which theoretically enables the saving of stream continuity. A second decision-making algorithm called a slow algorithm is furthermore provided to obtain an optimum distribution of the streams over the various remaining interfaces, after any reassigning of the stream or streams done by the fast algorithm. Furthermore, an appearance or re-appearance of an interface will be taken into account by the fast algorithm which could redirect streams traveling over other less advantageous interfaces towards this new interface, such a redirection being then followed by implementation of the slow algorithm in order to optimally distribute the stream over the various interfaces existing in the terminal considered.
However, the inventors have noted that the consecutive implementation of dual fast and slow algorithms does not provide for optimum operation of the telecommunications system, and more particularly of the terminal, in certain situations which will unfailingly occur when said system operates in real conditions. These particular conditions include one highly probable situation in which an interface does not disappear totally but is disturbed temporarily so that this interface no longer provides the streams designed to cross it with anything other than a total throughput rate that is very small when compared with the throughput rate initially planned for this interface. In such a situation, according to the teaching of the above-mentioned prior art, it is the slow algorithm that will have to take corrective measures in order to find a new optimum distribution of the streams: this entails a risk of introducing a period of latency with a duration that is big enough for the streams traveling through the saturated interface to be quite simply interrupted and lost, and this is unacceptable.